Abstract for Lisa Pearl's talk

Finding the right data to learn from: Computational modeling investigations
Lisa Pearl (University of Maryland)
March 6, 2007 (Tuesday)
6:30 PM - 8:00 PM; Room 7102, The CUNY Graduate Center

Learning language is a tricky business: the system is complex and the data are often noisy. It is vital to understand the mechanism of language learning given the boundary conditions provided by linguistic representation and the time course of acquisition. Moreover, it is not enough to restrict the potential systems that could be acquired, which can be done by defining a finite set of parameters the learner must set. Even supposing that the system is defined by n binary parameters, we must still explain how the learner converges on the correct system(s) out of the possible 2n systems, using data that is often highly ambiguous and exception-filled (Clark, 1994).

One potential solution is that learners filter the data used for learning (their data intake) down to a subset that is easier to extract systematicity from. However, this could lead to a data sparseness problem. If the learner can use only "good" data, are there enough "good" data to learn from?

Computational modeling proves itself a very useful tool in addressing this kind of question, since it would be difficult to explore with standard experimental techniques. In addition, the results of computational modeling can generate predictions that can then be tested experimentally.

The two computational modeling case studies presented will be embedded in a framework that is applicable to a range of language learning problems. In addition, this framework combines discrete linguistic representations with probabilistic methods and so can account for the gradualness and variation in learning that human children display. One of the discoveries from these case studies is that acquisition success seems to require that the data intake be a filtered subset of the available input. Moreover, filtering the data intake can lead to acquisition success even when the learner is faced with a complex, noisy system.